Wikimedia urges different fair use treatment for training and RAG

In its October 30, 2026 announcement, “The Wikimedia Foundation files to protect open knowledge in New York Times v OpenAI case,” published on its own website, the Wikimedia Foundation reports filing an amicus brief with the federal court for the Southern District of New York. The intervention concerns In re: OpenAI, Inc. Copyright Infringement Litigation, which combines twelve lawsuits, including the New York Times case against Microsoft and OpenAI. The central issue is the application of US fair use to copyrighted works used to develop language models. The filing and the brief’s contents have not been verified directly against the court document.

According to the announcement’s account, the New York Times alleges that use of its content, including for training, infringes copyright, while OpenAI and Microsoft invoke fair use for training and deploying models based on publicly available data. The Foundation proposes distinguishing between uses: training should qualify as fair use when it produces models that are new and different from the original works, without requiring compliance with their licensing terms. This is an argument submitted to the court, and whether it will be accepted remains to be determined.

For retrieval-augmented generation (RAG) and similar techniques, the Foundation instead argues that retrieving and copying news articles or other texts from the web to return them to users through chatbots should require compliance with applicable licences. For Wikipedia texts, it points to CC BY-SA 4.0, requiring attribution to authors and the same licence for adaptations. The proposed distinction seeks to preserve both tools that support open projects and the publishing sources on which knowledge available online depends.